Model training method and system, storage medium

By acquiring the meta-features of heterogeneous distributed semi-supervised datasets, determining the semi-supervised learning algorithm, and generating multiple models and their weights, the problems of model training efficiency and quality in heterogeneous distributed datasets in the telecommunications field are solved, achieving higher prediction accuracy and faster delivery speed.

CN114692886BActive Publication Date: 2026-06-05HUAWEI TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HUAWEI TECH CO LTD
Filing Date
2020-11-28
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

In machine learning applications in the telecommunications field, heterogeneous distributed datasets lead to inefficiencies and low quality in model delivery, and existing technologies struggle to effectively utilize unlabeled data for accurate model training.

Method used

By acquiring the meta-features of heterogeneous distributed semi-supervised datasets, a semi-supervised learning algorithm is determined. Based on margin density and statistical analysis parameters, multiple models and their weights are generated. The weighted information entropy minimization method is used to fuse the models, thereby improving the prediction accuracy of the models on heterogeneous distributed datasets.

Benefits of technology

It improves model accuracy in heterogeneous distributed semi-supervised dataset scenarios, enhances delivery efficiency and quality, and reduces trial-and-error costs and delivery cycles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114692886B_ABST
    Figure CN114692886B_ABST
Patent Text Reader

Abstract

Embodiments of the present application provide a model training method and system, and a storage medium, comprising: obtaining a heterogeneous distribution semi-supervised data set; determining a semi-supervised learning algorithm according to meta features of the heterogeneous distribution semi-supervised data set; obtaining M models and weights corresponding to the M models respectively according to the semi-supervised learning algorithm and the heterogeneous distribution semi-supervised data set, wherein the accuracy of the M models satisfies a preset condition. Through the embodiments of the present application, based on the heterogeneous distribution semi-supervised data set, the semi-supervised learning algorithm is determined according to the meta features of the heterogeneous distribution semi-supervised data set, and M models and weights corresponding to the M models respectively are obtained according to the semi-supervised learning algorithm and the heterogeneous distribution semi-supervised data set. By using this means, multiple models are obtained based on the heterogeneous distribution semi-supervised data set, the accuracy of the model obtained in the heterogeneous distribution semi-supervised data set scenario is improved, and the delivery efficiency and delivery quality are improved.
Need to check novelty before this filing date? Find Prior Art